Safety

Agent Skills by Addy Osmani 0.6.14 fixes unsafe samples

Agent Skills 0.6.14 fixes more than twenty code samples, including an unsafe rollback workflow, and stops Claude from starting /ship or /build on its own.

Date
Works with
Claude Code

What’s new

Agent Skills 0.6.14 came out on 29 October 2026. The release says: "More than twenty snippets across the skills either failed as written or taught the wrong thing, and they're fixed."

Two of the fixed samples were unsafe. The rollback workflow sample "passed its input straight into `run:`", which the release calls "a script injection in any repo that copied it". It now passes the input through env. The Lighthouse CI sample ran the unscoped lhci package, "which isn't Lighthouse CI". It now runs @lhci/cli.

Now only you can start the 9 slash commands, such as /spec, /build and /ship. The release says that without this "the model could start `/ship`, `/build` and the rest by itself". The constraints skill now scans the working tree for secrets, not only commits.

Why it matters

If you copied the old rollback workflow or the Lighthouse CI sample into your own project, check those files and change them to the new versions. Claude Code can no longer start a release step such as /ship by itself. You start these commands when you want them.

How to try it

Update the plugin with this command in your terminal: see the command below. Plugins from this marketplace do not update by themselves unless you turn on auto-update. Then start a new Claude Code session and type /spec to check that the commands are there.

claude plugin update agent-skills@addy-agent-skills

Source

Checked against the source on 29 Oct 2026

All updates